Real-time luminescence enables estimation of drug-induced proliferation rates in adherent and suspension cell lines

Required R packages

The diprate package is available from GitHub here: dipDRC

library(diprate)

Static data

static <- read.csv("../data/from_CW_via_Slack_20210507/StaticDF.csv", row.names=1)
static <- static[order(static$Culture_Type,static$Cell_Line),]
static[static$Cell_Conc==0,"Cell_Conc"] <- 1
cell_lines <- unique(static$Cell_Line)

Outlier value in CORL279

static <- static[!(static$Cell_Line=="CORL279" & static$RLU < 10),]

Code for reproducing figures

Correlation between luminescence and cell count

par(mfrow=c(2,5))
linear_models <- lapply(cell_lines, function(cl) {
        dat <- static[static$Cell_Line==cl,]
        culture_type <- unique(dat$Culture_Type)
        m <- lm(log2(RLU) ~ log2(Cell_Conc), data = dat)
        plot(log2(RLU) ~ log2(Cell_Conc), data=dat, main=paste0(cl," (",culture_type,")"), xlim=c(0,14), ylim=c(7,18))
        abline(m, col="blue")
        return(m)
    })

names(linear_models) <- cell_lines

Need 2-part function to accommodate minimum values

Assuming first part of data is at some minimum value and not associated with any actual cell count (lower limit of detection), which would result in slope=0 (values are constant until some minimum number of cells is achieved).

lagLine <- function (x, lower, slope, br = 64) sapply(x, function(z) ifelse(z <= br, lower, lower + (z-br) * slope))

fitLagLin <- function(x, y, start_list=list(lower=5, slope=1, br=1)) 
    nls(y ~ lagLine(x=x, lower, slope, br),
        start = start_list,
        algorithm="port",
        control=nls.control(maxiter=500)
        )

Fit the lag-linear model to data

par(mfrow=c(2,5))
lag_linear_models <- lapply(cell_lines, function(cl) {
    m <- tryCatch({
        dat <- static[static$Cell_Line==cl,]
        culture_type <- unique(dat$Culture_Type)
        m <- fitLagLin(log2(dat$Cell_Conc), log2(dat$RLU))
    }, error=function(e) { return(e) })
    # if(culture_type == "Adherent") 
    # {
    #     xr <- c(4,12)
    # } else {
    #     xr <- c(6,14)
    # }
    xr <- c(0,14)
    plot(log2(RLU) ~ log2(Cell_Conc), data=dat, main=paste0(cl," (",culture_type,")"), xlim=xr, ylim=c(7,18))
    if(class(m)[1] != "nls")
    {
        m <- lm(log2(RLU) ~ log2(Cell_Conc), data=dat[dat$Cell_Conc>1,])
        abline(m, col="blue", lwd=2)
    } else {
        curve(from=0.5,to=18, lagLine(x, lower=coef(m)['lower'], 
                                      slope=coef(m)['slope'], 
                                      br=coef(m)['br']), 
              col="blue", lwd=2, add=TRUE)
    }
    # text(12, 9, paste("Adj R2 ="))
    return(m)
})

names(lag_linear_models) <- cell_lines

Try eliminating controls

Assume all luminescence values with cells produce detectable signal.

par(mfrow=c(2,5))
linear_models <- lapply(cell_lines, function(cl) {
        dat <- static[static$Cell_Line==cl & static$Cell_Conc >1,]
        culture_type <- unique(dat$Culture_Type)
        m <- lm(log2(RLU) ~ log2(Cell_Conc), data = dat)
        plot(log2(RLU) ~ log2(Cell_Conc), data=dat, main=paste0(cl," (",culture_type,")"), xlim=c(0,14), ylim=c(7,18))
        abline(m, col="blue")
        return(m)
    })

names(linear_models) <- cell_lines

Compare to linear models

Assuming the lowest number of cells is above the threshold of detection, will remove the no cells control and fit remaining data.

dat <- static[static$Cell_Conc > 1,]
m2 <- lme4::lmList(log2(RLU) ~ log2(Cell_Conc) | Cell_Line, data=dat)
f2 <- coef(m2)
r2 <- unlist(summary(m2)$adj.r.squared)

par(mfrow=c(2,5))
temp <- lapply(cell_lines, function(cl) {
    dtp <- dat[dat$Cell_Line==cl,]
    culture_type <- unique(dtp[dtp$Cell_Line==cl,'Culture_Type'])
    if(culture_type == "Adherent") 
    {
        xr <- c(5,12)
    } else {
        xr <- c(7,14)
    }
    plot(log2(RLU) ~ log2(Cell_Conc), 
         data=dtp, 
         main=paste0(cl," (",culture_type,")"), 
         xlim=xr, ylim=c(7,18))
    abline(m2[[cl]], col="blue", lwd=2)
    text(xr[1]+0.25, 17, pos=4, paste("slope =",signif(f2[cl,2],3)))
    text(xr[1]+0.25, 16, pos=4, expression(R^2))
    text(xr[1]+1, 16, pos=4, paste("=", signif(r2[cl],3)))
})

fitLagLin1 <- function(x, y, start_list=list(lower=5, br=1)) 
    nls(y ~ lagLine(x=x, lower, slope=1, br),
        start = start_list,
        algorithm="port",
        control=nls.control(maxiter=500)
        )

par(mfrow=c(2,5))
lag_linear1_models <- lapply(cell_lines, function(cl) {
    m <- tryCatch({
        dat <- static[static$Cell_Line==cl,]
        culture_type <- unique(dat$Culture_Type)
        m <- fitLagLin1(log2(dat$Cell_Conc), log2(dat$RLU))
    }, error=function(e) { return(e) })
    # if(culture_type == "Adherent") 
    # {
    #     xr <- c(4,12)
    # } else {
    #     xr <- c(6,14)
    # }
    xr <- c(0,14)
    plot(log2(RLU) ~ log2(Cell_Conc), data=dat, main=paste0(cl," (",culture_type,")"), xlim=xr, ylim=c(7,18))
    if(class(m)[1] != "nls")
    {
        m <- lm(log2(RLU) ~ log2(Cell_Conc), data=dat[dat$Cell_Conc>1,])
        abline(m, col="blue", lwd=2)
    } else {
        curve(from=0.5,to=18, lagLine(x, lower=coef(m)['lower'], 
                                      slope=1, 
                                      br=coef(m)['br']), 
              col="blue", lwd=2, add=TRUE)
    }
    # text(12, 9, paste("Adj R2 ="))
    return(m)
})

names(lag_linear1_models) <- cell_lines

Combined cell count & lum data

lcc_cell_lines <- unique(lcc$cell.line)
ctrls <- lapply(lcc_cell_lines, function(cl) lcc[lcc$cell.line==cl & lcc$drug1.conc==0,])
names(ctrls) <- lcc_cell_lines

Cell counts

par(mfrow=c(2,2))
invisible(lapply(names(ctrls), function(n) do.call(plotGC, 
        append(getGCargs(ctrls[[n]], dat.col=c("time","Cell_Count","uid")),list(main=n, leg=FALSE)))))

Luminscence

par(mfrow=c(2,2))
invisible(lapply(names(ctrls), function(n) do.call(plotGC, 
        append(getGCargs(ctrls[[n]], dat.col=c("time","RLU","uid")),list(main=n)))))

dms53 <- ctrls[['DMS53']]

par(mfrow=c(1,2))

invisible(do.call(plotGC, append(getGCargs(dms53),list(main="DMS53, cell count", leg=FALSE))))
invisible(do.call(plotGC, append(getGCargs(dms53, dat.col=c("time","RLU","uid")),list(main="DMS53, lum", leg=FALSE))))

h1048 <- ctrls[['H1048']]

par(mfrow=c(1,2))

invisible(do.call(plotGC, append(getGCargs(h1048, dat.col=c("time","Cell_Count","uid")),list(main="H1048, cell count", leg=FALSE))))
invisible(do.call(plotGC, append(getGCargs(h1048, dat.col=c("time","RLU","uid")),list(main="H1048, lum", leg=FALSE))))

Sum of all control cell counts at each time point

invisible(do.call(plotGC, append(getGCargs(h1048, dat.col=c("time","RLU","uid")),list(main="H1048, lum", leg=FALSE))))
lines(log2(cell.count)-log2(cell.count)[1] ~ time, data=h1048_sumc, lwd=3)